Improve documentation for violin plot to illustrate per-gene usage – violin(adata.T, 'gene_property')

Hey! I thought plotting .var columns in sc.pl.violin() worked previously (and the error message seems to suggest this as well).

I am doing the following:

adata_counts.var['dropout_per_gene'] = (adata_counts.X > 0).mean(0)
adata_counts.obs['dropout_per_cell'] = (adata_counts.X > 0).mean(1)

sc.pl.violin(adata_counts, keys='dropout_per_gene')

and I get this error:

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-8-463060c90a0b> in <module>()
----> 1 sc.pl.violin(adata_counts, keys='dropout_per_gene')

~/scanpy/scanpy/plotting/anndata.py in violin(adata, keys, groupby, log, use_raw, stripplot, jitter, size, scale, order, multi_panel, show, xlabel, rotation, save, ax, **kwds)
    630                 X_col = adata.raw[:, key].X
    631             else:
--> 632                 X_col = adata[:, key].X
    633             obs_df[key] = X_col
    634     if groupby is None:

~/anndata/anndata/base.py in __getitem__(self, index)
   1303     def __getitem__(self, index):
   1304         """Returns a sliced view of the object."""
-> 1305         return self._getitem_view(index)
   1306 
   1307     def _getitem_view(self, index):

~/anndata/anndata/base.py in _getitem_view(self, index)
   1306 
   1307     def _getitem_view(self, index):
-> 1308         oidx, vidx = self._normalize_indices(index)
   1309         return AnnData(self, oidx=oidx, vidx=vidx, asview=True)
   1310 

~/anndata/anndata/base.py in _normalize_indices(self, index)
   1283         obs, var = super(AnnData, self)._unpack_index(index)
   1284         obs = _normalize_index(obs, self.obs_names)
-> 1285         var = _normalize_index(var, self.var_names)
   1286         return obs, var
   1287 

~/anndata/anndata/base.py in _normalize_index(index, names)
    261         return slice(start, stop, step)
    262     elif isinstance(index, (int, str)):
--> 263         return name_idx(index)
    264     elif isinstance(index, (Sequence, np.ndarray, pd.Index)):
    265         # here, we replaced the implementation based on name_idx with this

~/anndata/anndata/base.py in name_idx(i)
    248                 raise IndexError(
    249                     'Key "{}" is not valid observation/variable name/index.'
--> 250                     .format(i))
    251             i = i_found[0]
    252         return i

IndexError: Key "dropout_per_gene" is not valid observation/variable name/index.

The whole thing works for:

sc.pl.violin(adata_counts.T, keys='dropout_per_gene')
sc.pl.violin(adata_counts, keys='dropout_per_cell)

So it's clearly just not taking .var columns for sc.pl.violing().

I've also reproduced this with adata = sc.datasets.blob().

Start with the sc.pl.violin() entry point and reproduce the examples using adata_counts, adata_counts.T, and the blob dataset. Update the violin documentation to explain per-gene .var usage and the demonstrated working form, with the examples serving as the completion check.